LifeOS: A Life Assistant That Lives in Your Messages

Every personal-assistant app I've tried has the same failure mode: it wants to be opened. A dashboard, a home screen, a notification you swipe away and forget. LifeOS starts from the opposite assumption — the interface is a message thread you already have open, and the agent behind it does the opening, checking, and drafting so you don't have to.

Text it, and it reads your LinkedIn and Instagram activity through scoped, revocable access — not a standing firehose of everything, just the specific signals that matter: an upcoming birthday surfaced from a contact's profile, a post-worthy moment, a connection you haven't followed up with. It drafts the thing — a birthday post, a reply, a note — and holds it for approval instead of firing it off on your behalf.

The other half is the part most "AI assistant" demos skip: it keeps a running plan, not a to-do list. A to-do list is static until you open it again; LifeOS's plan updates itself as new signals come in, the same way a project plan gets revised mid-sprint instead of rewritten from scratch. Ask it what's next and it answers from the current state, not from whatever you typed in three weeks ago.

Underneath, it's the exact agent-lifecycle pattern I use in enterprise work, just pointed at a life instead of a bank: every connected account gets its own scoped identity, not a shared login; every draft is logged before it's sent, not after; and the plan-update loop is the same reviewer-and-writeback pattern that keeps production agents from drifting silently. Building it at this size is what lets me stress-test that pattern before it goes anywhere near 180,000 people.

It's a personal project, actively evolving — the next additions on the list are calendar-aware scheduling and a weekly digest instead of a constant stream. If message-driven, plan-not-list agents are something you're building toward too, this is the shape I'd start from.